Figure 5.

Structure of the simulated dataset. To perform statistical analyses within a controlled environment with pre-specified metabolite–outcome associations, we created a simulated dataset based generally on data features observed in multiple real-world datasets. One such simulated dataset demonstrates a scenario with multiple clusters of metabolites that have within-cluster correlation but little cross-cluster, mimicking the inter-relationships observed in actual experimentally derived human metabolomics studies [25,26,27].